[PATCH] Split out the memory handling from fio.c
[fio.git] / fio.c
diff --git a/fio.c b/fio.c
index 5432e383d439319eb918b3dd4162c2df774b0337..8d2a8e5f5ccc023ed32e8c16cccc72e683049890 100644 (file)
--- a/fio.c
+++ b/fio.c
@@ -45,8 +45,6 @@ int shm_id = 0;
 int temp_stall_ts;
 char *fio_inst_prefix = _INST_PREFIX;
 
-extern unsigned long long mlock_size;
-
 #define should_fsync(td)       ((td_write(td) || td_rw(td)) && (!(td)->odirect || (td)->override_sync))
 
 static volatile int startup_sem;
@@ -56,11 +54,10 @@ static volatile int startup_sem;
 
 static void terminate_threads(int group_id)
 {
+       struct thread_data *td;
        int i;
 
-       for (i = 0; i < thread_number; i++) {
-               struct thread_data *td = &threads[i];
-
+       for_each_td(td, i) {
                if (group_id == TERMINATE_ALL || groupid == td->groupid) {
                        td->terminate = 1;
                        td->start_delay = 0;
@@ -238,8 +235,6 @@ static int check_min_rate(struct thread_data *td, struct timeval *now)
                rate = (td->this_io_bytes[ddir] - td->rate_bytes) / spent;
                if (rate < td->ratemin) {
                        fprintf(f_out, "%s: min rate %d not met, got %ldKiB/sec\n", td->name, td->ratemin, rate);
-                       if (rate_quit)
-                               terminate_threads(td->groupid);
                        return 1;
                }
        }
@@ -277,7 +272,7 @@ static int get_rw_ddir(struct thread_data *td)
                 * Check if it's time to seed a new data direction.
                 */
                if (elapsed >= td->rwmixcycle) {
-                       int v;
+                       unsigned int v;
                        long r;
 
                        r = os_random_long(&td->rwmix_state);
@@ -422,7 +417,7 @@ static inline void td_set_runstate(struct thread_data *td, int runstate)
 
 static struct fio_file *get_next_file(struct thread_data *td)
 {
-       int old_next_file = td->next_file;
+       unsigned int old_next_file = td->next_file;
        struct fio_file *f;
 
        do {
@@ -688,14 +683,14 @@ static void do_io(struct thread_data *td)
        struct timeval s, e;
        unsigned long usec;
        struct fio_file *f;
-       int i;
+       int i, ret = 0;
 
        td_set_runstate(td, TD_RUNNING);
 
        while (td->this_io_bytes[td->ddir] < td->io_size) {
                struct timespec ts = { .tv_sec = 0, .tv_nsec = 0};
                struct timespec *timeout;
-               int ret, min_evts = 0;
+               int min_evts = 0;
                struct io_u *io_u;
 
                if (td->terminate)
@@ -728,9 +723,10 @@ static void do_io(struct thread_data *td)
                        min_evts = 1;
                }
 
+
                ret = td_io_getevents(td, min_evts, td->cur_depth, timeout);
                if (ret < 0) {
-                       td_verror(td, ret);
+                       td_verror(td, -ret);
                        break;
                } else if (!ret)
                        continue;
@@ -753,6 +749,8 @@ static void do_io(struct thread_data *td)
                rate_throttle(td, usec, icd.bytes_done[td->ddir]);
 
                if (check_min_rate(td, &e)) {
+                       if (rate_quit)
+                               terminate_threads(td->groupid);
                        td_verror(td, ENOMEM);
                        break;
                }
@@ -768,13 +766,15 @@ static void do_io(struct thread_data *td)
                        td_io_sync(td, f);
        }
 
-       if (td->cur_depth)
-               cleanup_pending_aio(td);
+       if (!ret) {
+               if (td->cur_depth)
+                       cleanup_pending_aio(td);
 
-       if (should_fsync(td) && td->end_fsync) {
-               td_set_runstate(td, TD_FSYNCING);
-               for_each_file(td, f, i)
-                       td_io_sync(td, f);
+               if (should_fsync(td) && td->end_fsync) {
+                       td_set_runstate(td, TD_FSYNCING);
+                       for_each_file(td, f, i)
+                               td_io_sync(td, f);
+               }
        }
 }
 
@@ -798,19 +798,7 @@ static void cleanup_io_u(struct thread_data *td)
                free(io_u);
        }
 
-       if (td->mem_type == MEM_MALLOC)
-               free(td->orig_buffer);
-       else if (td->mem_type == MEM_SHM) {
-               struct shmid_ds sbuf;
-
-               shmdt(td->orig_buffer);
-               shmctl(td->shm_id, IPC_RMID, &sbuf);
-       } else if (td->mem_type == MEM_MMAP)
-               munmap(td->orig_buffer, td->orig_buffer_size);
-       else
-               log_err("Bad memory type %d\n", td->mem_type);
-
-       td->orig_buffer = NULL;
+       free_io_mem(td);
 }
 
 static int init_io_u(struct thread_data *td)
@@ -829,32 +817,8 @@ static int init_io_u(struct thread_data *td)
 
        td->orig_buffer_size = td->max_bs * max_units + MASK;
 
-       if (td->mem_type == MEM_MALLOC)
-               td->orig_buffer = malloc(td->orig_buffer_size);
-       else if (td->mem_type == MEM_SHM) {
-               td->shm_id = shmget(IPC_PRIVATE, td->orig_buffer_size, IPC_CREAT | 0600);
-               if (td->shm_id < 0) {
-                       td_verror(td, errno);
-                       perror("shmget");
-                       return 1;
-               }
-
-               td->orig_buffer = shmat(td->shm_id, NULL, 0);
-               if (td->orig_buffer == (void *) -1) {
-                       td_verror(td, errno);
-                       perror("shmat");
-                       td->orig_buffer = NULL;
-                       return 1;
-               }
-       } else if (td->mem_type == MEM_MMAP) {
-               td->orig_buffer = mmap(NULL, td->orig_buffer_size, PROT_READ | PROT_WRITE, MAP_PRIVATE | OS_MAP_ANON, 0, 0);
-               if (td->orig_buffer == MAP_FAILED) {
-                       td_verror(td, errno);
-                       perror("mmap");
-                       td->orig_buffer = NULL;
-                       return 1;
-               }
-       }
+       if (allocate_io_mem(td))
+               return 1;
 
        p = ALIGN(td->orig_buffer);
        for (i = 0; i < max_units; i++) {
@@ -1087,15 +1051,19 @@ static void *fork_main(int shmid, int offset)
  */
 static void reap_threads(int *nr_running, int *t_rate, int *m_rate)
 {
+       struct thread_data *td;
        int i, cputhreads;
 
        /*
         * reap exited threads (TD_EXITED -> TD_REAPED)
         */
-       for (i = 0, cputhreads = 0; i < thread_number; i++) {
-               struct thread_data *td = &threads[i];
-
-               if (td->io_ops->flags & FIO_CPUIO)
+       cputhreads = 0;
+       for_each_td(td, i) {
+               /*
+                * ->io_ops is NULL for a thread that has closed its
+                * io engine
+                */
+               if (td->io_ops && td->io_ops->flags & FIO_CPUIO)
                        cputhreads++;
 
                if (td->runstate != TD_EXITED)
@@ -1120,48 +1088,6 @@ static void reap_threads(int *nr_running, int *t_rate, int *m_rate)
                terminate_threads(TERMINATE_ALL);
 }
 
-static void fio_unpin_memory(void *pinned)
-{
-       if (pinned) {
-               if (munlock(pinned, mlock_size) < 0)
-                       perror("munlock");
-               munmap(pinned, mlock_size);
-       }
-}
-
-static void *fio_pin_memory(void)
-{
-       unsigned long long phys_mem;
-       void *ptr;
-
-       if (!mlock_size)
-               return NULL;
-
-       /*
-        * Don't allow mlock of more than real_mem-128MB
-        */
-       phys_mem = os_phys_mem();
-       if (phys_mem) {
-               if ((mlock_size + 128 * 1024 * 1024) > phys_mem) {
-                       mlock_size = phys_mem - 128 * 1024 * 1024;
-                       fprintf(f_out, "fio: limiting mlocked memory to %lluMiB\n", mlock_size >> 20);
-               }
-       }
-
-       ptr = mmap(NULL, mlock_size, PROT_READ | PROT_WRITE, MAP_PRIVATE | OS_MAP_ANON, 0, 0);
-       if (!ptr) {
-               perror("malloc locked mem");
-               return NULL;
-       }
-       if (mlock(ptr, mlock_size) < 0) {
-               munmap(ptr, mlock_size);
-               perror("mlock");
-               return NULL;
-       }
-
-       return ptr;
-}
-
 /*
  * Main function for kicking off and reaping jobs, as needed.
  */
@@ -1170,9 +1096,9 @@ static void run_threads(void)
        struct thread_data *td;
        unsigned long spent;
        int i, todo, nr_running, m_rate, t_rate, nr_started;
-       void *mlocked_mem;
 
-       mlocked_mem = fio_pin_memory();
+       if (fio_pin_memory())
+               return;
 
        if (!terse_output) {
                printf("Starting %d thread%s\n", thread_number, thread_number > 1 ? "s" : "");
@@ -1187,9 +1113,7 @@ static void run_threads(void)
        nr_started = 0;
        m_rate = t_rate = 0;
 
-       for (i = 0; i < thread_number; i++) {
-               td = &threads[i];
-
+       for_each_td(td, i) {
                print_status_init(td->thread_number - 1);
 
                init_disk_util(td);
@@ -1218,9 +1142,7 @@ static void run_threads(void)
                /*
                 * create threads (TD_NOT_CREATED -> TD_CREATED)
                 */
-               for (i = 0; i < thread_number; i++) {
-                       td = &threads[i];
-
+               for_each_td(td, i) {
                        if (td->runstate != TD_NOT_CREATED)
                                continue;
 
@@ -1309,9 +1231,7 @@ static void run_threads(void)
                /*
                 * start created threads (TD_INITIALIZED -> TD_RUNNING).
                 */
-               for (i = 0; i < thread_number; i++) {
-                       td = &threads[i];
-
+               for_each_td(td, i) {
                        if (td->runstate != TD_INITIALIZED)
                                continue;
 
@@ -1336,7 +1256,7 @@ static void run_threads(void)
        }
 
        update_io_ticks();
-       fio_unpin_memory(mlocked_mem);
+       fio_unpin_memory();
 }
 
 int main(int argc, char *argv[])